[avr-libc-dev] Re: [bug #29964] Possible Incorrect Interrupt Vector Addresses for mega168

[Jose Torres]

Hi Joerg,

Consider it my first patch proposal. :)

- Jose

On Wed, May 26, 2010 at 1:17 AM, Joerg Wunsch <address@hidden>wrote:

>
> Update of bug #29964 (project avr-libc):
>
>                  Status:               Need Info => Invalid
>             Open/Closed:                    Open => Closed
>
>    _______________________________________________________
>
> Follow-up Comment #3:
>
> Hi Jose,
>
> for a simple test, you could just compile and link an empty main(),
> and run it through the disassembler.  That will show you the vector
> table is fine for both devices.  Btw., be assured if they weren't, the
> Arduino folks would have told us long before. ;-)
>
> > Are the addresses properly computed because of the following?
>
> _VECTORS_SIZE is part of the story: if you look into gcrt1.S, you'll
> notice it determines how many vectors will actually be emitted.
>
> The size logic itself is hidden behind the XJMP macro which comes from
> common/macros.inc.  It uses the (misnamed) macro __AVR_MEGA__ which is
> set by the compiler to determine whether to emit a JMP or RJMP
> instruction (which in turn determines the size of a vector table
> slot).  This should use __AVR_HAVE_JMP_CALL__ instead, but this one is
> only available starting with GCC 4.3, and we'd like to be backwards
> compatible.  We should perhaps check for the GCC version, and then use
> wither __AVR_MEGA__ or __AVR_HAVE_JMP_CALL__, depending on the
> version, so __AVR_MEGA__ could eventually be retired.  Feel free to
> submit a patch for this if you like (preferrably, tested against
> different compiler versions... ;-).
>
>
>    _______________________________________________________
>
> Reply to this item at:
>
>  <http://savannah.nongnu.org/bugs/?29964>
>
> _______________________________________________
>  Message sent via/by Savannah
>  http://savannah.nongnu.org/
>
>